Claude Drops the Wall of Text, and the AI Ads War Goes Public

Claude stops giving you a wall of text

The most practical change this week is one every Claude user will hit. Claude is actively moving away from the classic wall of text and toward interactive interfaces. Ask for the best ramen near you and it fetches your area, then returns a live map you can pan around, with a directions button that hands you off to Google Maps. It is not only maps. There is a weather widget, and recipe queries now return a custom interface with images and a dedicated measurements button. Crucially, this is available to both free and paid users.

The host's read is that this is just the start. Right now Claude covers the main categories people search for, but the logical end state is a tailored interface for every result, an app suited to the question rather than a default block of text. A year from now this could look very different.

Codex 5.3 versus Opus 4.6

On the coding side, OpenAI's Codex is getting scary good at building applications. The episode compared Codex 5.3 and Opus 4.6 on two prompts:

• A Death Star over LA in SVG, where Opus produced the first version you could actually recognize as LA, with palm trees and a skyline
• An ocean wave simulator, where Opus delivered a 3D scene with light reflections, a movable sun, a sunset feel, and a stormy preset

- A Death Star over LA in SVG, where Opus produced the first version you could actually recognize as LA, with palm trees and a skyline - An ocean wave simulator, where Opus delivered a 3D scene with light reflections, a movable sun, a sunset feel, and a stormy preset

But one-off demos do not tell the whole story. Codex 5.3 is 25% faster than 5.2 and can work for days without losing context, which matters for real developer workflows on existing codebases. The takeaway is to pick by use case: Opus for quick, pretty, one-off builds, Codex for sustained developer work.

The ads war breaks into the open

The bigger cultural story is ads. OpenAI started rolling out ads, but only for free US accounts, kept clearly separated from results. Competitors do not show ads yet. Anthropic did not hesitate and ran ads poking fun at OpenAI that were genuinely funny, enough that Sam Altman tweeted they were funny even as he argued they did not represent OpenAI's approach fairly.

AI has become so mainstream that it was the main topic at the Super Bowl, where it felt like most ads were about AI or made by AI.

If you pay, none of this touches you yet. But most people are on free accounts, so this will shape the mainstream experience of AI.

The quick hits that matter

A few more releases are worth knowing. ChatGPT Deep Research now runs on GPT-5.2 with longer reports, a side table of contents, and the ability to target specific sites, feeling closer to Gemini's version. Perplexity model council sends your question to OpenAI, Anthropic, and Gemini and returns a combined answer, automating a habit power users already have. On video, Seedance 2.0 drew heavy hype and Kling 3.0 generates 15-second multi-cut product videos with strong character consistency. Finally, Eleven Labs shipped a lifelike audiobook toolkit and a more expressive mode for its voice agents. Taken together, the week points one direction: AI is moving from clever text to tailored interfaces and real consumer products.